Research On Assemblyand Matching Algorithm Of Shaft And Hole Parts Based On 3D Point Cloud Map

The shaft and hole matching is the interaction of the of the shaft and hole contact surfaces.As a result of machining errors in part processing,the accuracy and performance after matching are also greatly different.When the fitting parts are asked to seal good and easily dismantle,it is required that the shape and size of the fitting surface of shaft and hole parts should be as similar as possible,the fitting surfaces should fit evenly,and the fitting distance at each point should be close to zero.The traditional method mostly uses vernier caliper to measure the actual size of multiple points on the fitting surface of shaft and hole parts,but this method can only measure the actual size of a limited number of points on the fitting surface,which lead to larger errors in the matching results,and the manual measurement is time-consuming and laborious.In this dissertation,three-dimensional point cloud map is used to solve the assembly and matching problem of shaft and hole parts.3D point cloud map is got by scanning the original parts and extracting a lot of points from the surface.The point cloud can express the structure,size and other characteristics of the original model.This dissertation mainly includes the following contents:The 3D point cloud map of the part is obtained by scanning the original part,in this dissertation,a series of pre-processing,such as filtering and noise reduction,topological relation establishment,boundary feature point extraction,non-boundary feature point simplification,and 3D reconstruction,are implemented to increase the accuracy of 3D point cloud and thus a more precise reflection on the original features.Because of the missing holes in the original point cloud,this dissertation uses the depth map and the convolutional autoencoder neural network to complete the hole filling of the point cloud with holes through supervised learning.Due to the surface bump of parts or light blocking in scanning,there are inevitably holes in the 3D point cloud map,which greatly affects the accuracy of the feature vector.To solve this problem,this paper uses the depth map and the convolutional autoencoder neural network to complete the hole filling of the point cloud with holes through supervised learning.The algorithm can be used to fill the simple holes like circle or square on the cylindrical surface.In this dissertation,based on point cloud feature operator,an assembly and matching algorithm of shaft and hole parts is proposed.The Euclidean distance between feature operators of parts is taken as the standard to measure the fitting degree of parts.Based on ICP registration and K-D tree,the shaft and hole parts are virtually assembled,the best fitting position is figured out,and the assembly clearance of parts is calculated.Based on Visual Studio and Qt creator,the application software is developed,which can complete a series of functions,such as point cloud data preprocessing,point cloud feature operator extraction,shaft hole parts assembling and matching,fit status analysis and so on.